The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
Filtros de partículas têm sido amplamente utilizados para problemas de estimativa de estado em sistemas não lineares e não gaussianos. Seu desempenho depende do sistema fornecido e dos modelos de medição, que precisam ser projetados pelo usuário para cada sistema alvo. Este artigo propõe um novo método para projetar esses modelos para um filtro de partículas. Este é um método de otimização numérica, onde o processo de projeto do filtro de partículas é interpretado no âmbito da aprendizagem por reforço, atribuindo as aleatoriedades incluídas em ambos os modelos do filtro de partículas à política de aprendizagem por reforço. Neste método, a estimação pelo filtro de partículas é realizada repetidamente e os parâmetros que determinam ambos os modelos são atualizados gradativamente de acordo com os resultados da estimação. A vantagem é que ele pode otimizar diversas funções objetivo, como a precisão da estimativa do filtro de partículas, a variância das partículas, a probabilidade dos parâmetros e o termo de regularização dos parâmetros. Derivamos as condições para garantir que o cálculo de otimização converge com a probabilidade 1. Além disso, para mostrar que o método proposto pode ser aplicado a problemas de escala prática, projetamos o filtro de partículas para localização de robôs móveis, que é uma tecnologia essencial para navegação autônoma. Por simulações numéricas, demonstra-se que o método proposto melhora ainda mais a precisão da localização em comparação ao método convencional.
Ryota YOSHIMURA
Kyoto University,Tokyo Metropolitan Industrial Technology Research Institute
Ichiro MARUTA
Kyoto University
Kenji FUJIMOTO
Kyoto University
Ken SATO
Tokyo Metropolitan Industrial Technology Research Institute
Yusuke KOBAYASHI
Tokyo Metropolitan Industrial Technology Research Institute
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copiar
Ryota YOSHIMURA, Ichiro MARUTA, Kenji FUJIMOTO, Ken SATO, Yusuke KOBAYASHI, "Particle Filter Design Based on Reinforcement Learning and Its Application to Mobile Robot Localization" in IEICE TRANSACTIONS on Information,
vol. E105-D, no. 5, pp. 1010-1023, May 2022, doi: 10.1587/transinf.2021EDP7192.
Abstract: Particle filters have been widely used for state estimation problems in nonlinear and non-Gaussian systems. Their performance depends on the given system and measurement models, which need to be designed by the user for each target system. This paper proposes a novel method to design these models for a particle filter. This is a numerical optimization method, where the particle filter design process is interpreted into the framework of reinforcement learning by assigning the randomnesses included in both models of the particle filter to the policy of reinforcement learning. In this method, estimation by the particle filter is repeatedly performed and the parameters that determine both models are gradually updated according to the estimation results. The advantage is that it can optimize various objective functions, such as the estimation accuracy of the particle filter, the variance of the particles, the likelihood of the parameters, and the regularization term of the parameters. We derive the conditions to guarantee that the optimization calculation converges with probability 1. Furthermore, in order to show that the proposed method can be applied to practical-scale problems, we design the particle filter for mobile robot localization, which is an essential technology for autonomous navigation. By numerical simulations, it is demonstrated that the proposed method further improves the localization accuracy compared to the conventional method.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2021EDP7192/_p
Copiar
@ARTICLE{e105-d_5_1010,
author={Ryota YOSHIMURA, Ichiro MARUTA, Kenji FUJIMOTO, Ken SATO, Yusuke KOBAYASHI, },
journal={IEICE TRANSACTIONS on Information},
title={Particle Filter Design Based on Reinforcement Learning and Its Application to Mobile Robot Localization},
year={2022},
volume={E105-D},
number={5},
pages={1010-1023},
abstract={Particle filters have been widely used for state estimation problems in nonlinear and non-Gaussian systems. Their performance depends on the given system and measurement models, which need to be designed by the user for each target system. This paper proposes a novel method to design these models for a particle filter. This is a numerical optimization method, where the particle filter design process is interpreted into the framework of reinforcement learning by assigning the randomnesses included in both models of the particle filter to the policy of reinforcement learning. In this method, estimation by the particle filter is repeatedly performed and the parameters that determine both models are gradually updated according to the estimation results. The advantage is that it can optimize various objective functions, such as the estimation accuracy of the particle filter, the variance of the particles, the likelihood of the parameters, and the regularization term of the parameters. We derive the conditions to guarantee that the optimization calculation converges with probability 1. Furthermore, in order to show that the proposed method can be applied to practical-scale problems, we design the particle filter for mobile robot localization, which is an essential technology for autonomous navigation. By numerical simulations, it is demonstrated that the proposed method further improves the localization accuracy compared to the conventional method.},
keywords={},
doi={10.1587/transinf.2021EDP7192},
ISSN={1745-1361},
month={May},}
Copiar
TY - JOUR
TI - Particle Filter Design Based on Reinforcement Learning and Its Application to Mobile Robot Localization
T2 - IEICE TRANSACTIONS on Information
SP - 1010
EP - 1023
AU - Ryota YOSHIMURA
AU - Ichiro MARUTA
AU - Kenji FUJIMOTO
AU - Ken SATO
AU - Yusuke KOBAYASHI
PY - 2022
DO - 10.1587/transinf.2021EDP7192
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E105-D
IS - 5
JA - IEICE TRANSACTIONS on Information
Y1 - May 2022
AB - Particle filters have been widely used for state estimation problems in nonlinear and non-Gaussian systems. Their performance depends on the given system and measurement models, which need to be designed by the user for each target system. This paper proposes a novel method to design these models for a particle filter. This is a numerical optimization method, where the particle filter design process is interpreted into the framework of reinforcement learning by assigning the randomnesses included in both models of the particle filter to the policy of reinforcement learning. In this method, estimation by the particle filter is repeatedly performed and the parameters that determine both models are gradually updated according to the estimation results. The advantage is that it can optimize various objective functions, such as the estimation accuracy of the particle filter, the variance of the particles, the likelihood of the parameters, and the regularization term of the parameters. We derive the conditions to guarantee that the optimization calculation converges with probability 1. Furthermore, in order to show that the proposed method can be applied to practical-scale problems, we design the particle filter for mobile robot localization, which is an essential technology for autonomous navigation. By numerical simulations, it is demonstrated that the proposed method further improves the localization accuracy compared to the conventional method.
ER -